How a cache implementation should be created depends on whether the caches can/should be declared ahead of time. For caches with predefined cache names, SyncCache beans should be created which will be ingested by the default cache manager. For caches that cannot or should not be defined ahead of time, a DynamicCacheManager bean should be registered to retrieve the cache.
Any other beans created are not used by Micronaut. The only beans referenced are SyncCache and DynamicCacheManager.
For caches that using blocking IO, the getExecutorService() method should be overridden for each SyncCache to allow the offloading of the operations to a thread pool. Unless there is a reason to do otherwise, the TaskManagers.IO named thread pool should be used.
If the cache provider has an asynchronous API, the async() method of SyncCache should be overridden to supply an AsyncCache that uses the native asynchronous API. See the Hazelcast implementation for an example.
To ensure consistency across implementations, copy the build of an existing implementation and alter the dependencies as needed.
All cache implementations should be documented by modifying the src/main/docs/guide/toc.yml to include your documentation. Documentation should describe how to install and configure the implementation.